AX Engine 7.5.x ships dense 27B past the DRAM ceiling.See the numbers

AX ENGINE / Installation guide

Install AX Engine.

Homebrew is the primary channel on macOS 26+ Apple Silicon. Python (venv) is the secondary channel for SDK consumers. Both ship the same server and bench tools.

01 / Install

1. Install

Pick a channel and run the command below in Terminal on macOS 26+ Apple Silicon.

macOS 26+ Apple SiliconTerminal

Open source · Apache-2.0

Install the latest public release from GitHub releases. v7.5.3 is the current line on Homebrew `defai-digital/tap/ax-engine`.

brew install defai-digital/tap/ax-engine

Homebrew installs the standalone CLI, the server, and the bench binary without Xcode or a Python toolchain.

02 / AX ENGINE

2. Run the doctor

Run `ax-engine doctor` to verify the bundle, MLX metallib, and HF cache layout.

ax-engine doctor

03 / AX ENGINE

3. Serve the default pack

AX Engine downloads the pinned Qwen 3.8 27B AXQ 6-bit MTP pack the first time you serve it. The default listen address is 127.0.0.1:31418.

ax-engine serve qwen3.8-27b:axq

04 / OpenAI

4. Send your first request

Use any OpenAI-compatible client. The same endpoint serves chat, embeddings, audio, and vision depending on the model.

curl http://127.0.0.1:31418/v1/chat/completions -H 'content-type: application/json' -d '{"model":"qwen3.8-27b","messages":[{"role":"user","content":"Say hello in one sentence."}],"max_tokens":64}'